Research On The Real-Time Attitude And Position Measuring System Based On Stereo Vision

With the development of the computer vision theory and vision sensors, the stereo vision is widely used in manufacture, aeronautics and robots navigation. This thesis presents a real-time method of measuring the attitude and position of a moving object based on binocular stereo vision. The researching contents include four aspects: schematic design and the basic principle of measuring system, feature extraction, stereo matching and token tracking.At first, the overall plan of the real-time measuring system based on stereo vision was introduced in this paper. After the basic principle of the measuring system was represented, the shemes of dual cameras synchronization and real-time data processing were designed. Based on the coordinates of feature points calculated by using intersection principle, an algorithm of calculating the relative attitude and position of moving target was derived.Sencondary, the methods of feature extraction were studied. For the X-junction in chessboard image, two different kinds of subpixel detection methods were studied and contrasted. For the features of the moving object, the center of the facula feature was extracted. The corner feature was extracted by using the method combining the gray center of the gravity and the distribution of the corner respond function.Then, the methods of stereo matching and token tracking were studied. The task of stereo matching is that determine correspondence and transformation between two sets of feature points in the left and right camera. It is the significant problems to reconstruct the 3D feature in stereo vision. The method based on epipolar geometry of stereo vision system is deeply discussed. The method present a fast way to setup the matching sets between two images. With the exclusive and sequence consistent restriction, the corresponding of the feature points between the left and right iamge was accomplished. Otherwise, an algorithm of token tracking was researched. Using the image coordinates of feature and trajectories of the points, the same physical feature point in a sequence image was identified. Considering the situation that the feature point may disappear in a few frames, a method used to predict the missed feature point was presented.Finally, a whole processing software was developed. The software consists of two softwares which have different effects. The digital video recording software was mainly used for recording real-time image sequences to hard disk subsystem and real-time image processing. The software of calculating the attitude and postion was used to receive the object data from net and calculate the attitude and postion of the moving object. A new method which was used to estimate the relationship between two different coordinate frames was introduced. After configuring the whole measuring system and debugging successfully, an experiment of the whole measurement system was carried on, and the experiment data was analyzed particularly.